A Statistical Analysis of the Higgs Boson
Abstract: The writing begins with basic statistics, null hypothesis, signal-free datasets and
move on to more advanced statistical analysis. The final report scrutinizes the standard
model H in pp collisions at s = 13 TeV using pseudo-experiment training datasets to
optimize event selection.
Shouyang Michael Wang
University of Washington
Physics 434 Capstone Project
Advisor: Dr. Miguel Morales
December 2019
2
Contents
Background ................................................................................................................................................... 3
Statistics ........................................................................................................................................................ 5
Normal Distribution ...................................................................................................................................... 5
χ2-distribution ............................................................................................................................................... 6
Poisson distribution ...................................................................................................................................... 7
Central Limit Theorem .................................................................................................................................. 8
Multidimensional Gaussian ........................................................................................................................... 9
Higgs Bosons & QCD Parameters ................................................................................................................ 11
DATA EXPLORATION.................................................................................................................................... 12
TRANSVERSE MOMENTUM(PT) (XY PLANE)(UNIT: GEV*C-1 OR GEV) ........................................................ 13
PSEUDORAPIDITY ETA (η)............................................................................................................................ 14
PHI (Φ) ......................................................................................................................................................... 15
INVARIANT MASS (M)(UNIT: GEV*C-2 OR GEV) ......................................................................................... 16
EE2(E2) AND EE3(E3): ENERGY CORRELATION FUNCTIONS ........................................................................ 18
D2(D2) JET SHAPE FUNCTION, USED IN TAGGING ...................................................................................... 20
ANGULARITY ............................................................................................................................................... 21
N-SUBJETTINESS(𝜏N); JET SHAPE FUNCTION, USED IN TAGGING ............................................................... 22
KTDELTAR(KT ΔR) KT CLUSTERING ........................................................................................................... 24
Higgs Sensitivity .......................................................................................................................................... 25
Conclusions ................................................................................................................................................. 27
Bibliography ................................................................................................................................................ 28
3
Background
The Standard Model
The Standard Model of particle physics describes all known fundamental particles that are
the building blocks to everything in the universe.
The Higgs mechanism, a mechanism theorized to give rise to the masses of all the
subatomic elements, is essential for the Standard Model to work mathematically - all
particles that mediate forces are massless. The mechanism successfully explained how
particles obtain their mass through the Higgs field and theorized the Higgs bosons
existence.
Higgs boson (H) is one of the elementary particles in the Standard Model. The particle is
extremely unstable, meaning that it would decay into other particles almost immediately.
Higgs boson has been verified experimentally using the Large Hadron Collider at the
European Organization for Nuclear Research(CERN).
Large Hadron Collider
The Large Hadron Collider is a circular high-energy particle accelerator and collider. It
consists of a ring of “superconducting magnets,” along with several accelerating structures.
Inside the ring, two beams containing protons or lead ions would travel in opposite
directions to collide at speed close to the speed of light. Particle detectors are placed
around the collision site to record and analyze the subatomic particles that arise from the
collision. The LHC is engineered to operate at a maximum collision energy of 14 TeV,
meaning that each traveling beam would carry a power of 7 TeV. However, in most
operations, the collision energy is controlled under 14 TeV to “optimize the delivery of
particle collisions” (CERN).
Furthermore, one of the particle detectors is named ATLAS(A Toroidal LHC ApparatuS). It
is used to detect a wide range of subatomic particles, including the Higgs boson.
(Source: CERN)
4
Collision Energy
In the context of proton-proton(pp) collisions, √𝑠 indicates the total center-of-mass energy
of the colliding beams.
In the laboratory frame, the center-of-mass energy is calculated as the following:
We will observe and analyze an inclusive search for the standard model Higgs boson in
typical pp collisions at √𝑠 = 13 TeV at the LHC using simulated data. In other words, the
energy for one beam of protons is 6.5 TeV, which would result in the total energy of √𝑠 =13
TeV when two beams collide.
Jets
The term refers to a large number of detectable particles flying roughly in the same
direction. The process is called clustering.
The decay of Higgs boson
The Higgs boson is extremely unstable, and it would decay into other particles almost
immediately. The standard model of particle physics theorized that the Higgs could decay
into other particles like quarks. There are many “flavors” when it comes to quarks. The net
electric charge of the byproducts of the decay has to be zero since the Higgs boson has a
charge of zero. Quarks have electric charges, meaning that the quarks created from the
decay have to emerge in pairs, such that the charge from one quark and one antiquark
would cancel each other out, resulting in a net charge of zero. Very often, physicists find
that the Higgs boson would decay into a pair of bottom quarks.
Quantum Chromodynamics(QCD)
During a proton-proton(pp) collision, almost all detected events are due to the strong
interaction(QCD). Due to the Higgs bosons unstable nature, the pp collision would lead to a
pair of “particle jets originating from b quarks fragmentation. (ATLAS)” It is incredibly
challenging to distinguish an actual Higgs signal from the” unwanted” b-quark pairs noise
due to the QCD background.
(Source: Particle Physics Theory, University of Glasgow)
5
Statistics
Normal Distribution
In physics, σ refers to a probability. The sigma implicitly refers to the standard normal
distribution (a Gaussian with mean zero and a standard deviation of 1).
Integrals of the standard normal distribution give probabilities. The normal distribution is
a symmetric distribution where most of the observations cluster around the center to form
a bell shape. A Gaussian would remain a Gaussian after summation, multiplication, or
Fourier transform.
The z-score is positive if the value lies above the mean and negative if it lies below the
mean. Notice that it is possible to get a “negative value” when integrating from a negative
number to another number.
For instance, if we try to integrate a function from negative infinity to a number, it’s
possible to get a negative area. It does not make any sense to have a negative probability.
The probability from -10 to 0 should have the same probability from 0 to 10 for a Gaussian
distribution since the distribution is symmetrical.
To illustrate this, we try to integrate the equation for the normal distribution using two
different sets of limits.
Notice that the two equations generate the same probability. In other words, the
area(probability) under the curve from -10 to 0 is the same as the area from 0 to 10.
6
χ2-distribution
Now we explore some other continuous analytic distributions. The χ2-distribution (or Chi-
squared) with k degrees of freedom is the distribution of a sum of squares of k independent
standard normal random variables. Now we create a χ2-distribution object with the
parameter value b = 2, a special case of Gamma distribution.
Assuming that we have signal-free data that follows the χ2-distribution, and we have a
measurement for which we need to determine the σ. The test statistic is measured to be
7.1. We propose a statistical question: What is the probability that the background
produces a signal that is less than the measurement we made? More specifically, given the
test statistics and if we know the distribution has 3 degrees of freedom, what is the area to
the left of the test statistic? First, we try to approach the question mathematically. The
formula for the cumulative distribution function of the chi-square distribution is given by
7
Poisson distribution
Now we move on to non-continuous distributions. We will be looking at the Poisson
distribution.
Notice that
1) all of the variables are discrete.
2) The graphs have a lower bond at zero(For instance, the data cant go lower than 0), but
they dont have an upper limit.
3)The graph with a low mean(lambda) is highly skewed. 4) the distribution becomes more
bell-like as the mean is getting large.
5) The shape isnt very symmetric like the normal distribution.
6) The mean, lambda, is a constant number throughout the experiment.
We propose a statistical question: Given that lambda = 3, what is the probability of the
background producing a precisely 4? It can be calculated as the following:
Because the distributions are discrete, so are the probabilities and σ. A discrete number has
countable values. Since the distribution is discrete, it makes sense to talk about probability
at an exact value like P(x=4) shown above. The discrete nature is more pragmatic when it
comes to experiments that expect a specified number of events to take place.
While the results are discrete, the parameters of the distributions are not. For example, the
mean of a Poisson distribution can be 9.2. The mean is essentially the average of a set of
numbers. It is an arithmetic number obtained from experimental data. It does not have to
be a whole number. However, since the distribution is typically used to reflect the
probability of a number of events occurring in a fixed interval, it does not make much sense
to say that an event happened 1.5 times. An event can happen on average 1.5 times.
Typically the x-axis represents a given interval of time(e.g., how many times an event
8
happens at hours 1, 2,3,4, etc.). Just like its either head or tail, it cant be somewhere in
between.
Central Limit Theorem
Furthermore, the sum of a Poisson with itself still has Poisson distribution. To prove this,
we will use the moment generating function of the Poisson distribution. It is given by
For a linear combination,
Hence,
LaTex code:
From a conceptual point of view, the Poisson distribution is the number of counts in a fixed
interval of time. When we conv() or sum a Poisson distribution with itself N(in this case,
five) times, it is the same as saying that we multiply the interval by N, or N times longer
than the original interval. Consequently, the mean is also multiplied by N. The two methods
(sum N times or multiply interval and mean by N)would get the same result. A Poisson
distribution with a rescaled interval and mean is still a Poisson distribution.
A limiting form of the Poisson distribution is the Gaussian distribution. The central limit
theorem states that when independent random variables are summed or averaged enough
trials, their normalized sum or average tends toward a normal distribution. Poisson is
becoming more symmetrical and bell-like as it is averaging over time. Because of the
central limit theorem, Poisson tends to normal as its mean becomes sufficiently large.
9
Multidimensional Gaussian
10
11
Higgs Bosons & QCD Parameters
12
DATA EXPLORATION
The Higgs bosons are produced with large transverse momentum (pT) and decay to a
bottom quark-antiquark pair. The Higgs candidates could be reconstructed as large-radius
jets using Calorimeters. Due to large QCD background contamination, the direct 5-sigma
observation of this Higgs channel is not accomplished yet[Phys. Rev. Lett. 120, 071802
(2018)]. We will analyze the data using MATLAB. See lab5.m for the complete code. The
signal dataset is labeled as “Higgs,” and the background dataset is labeled as “QCD.” First,
we import the data. In the data, the transverse momentum ranges from 250 GeV to 500
GeV.
As shown above, there are two datasets. The first dataset is the Higgs signal, and the second
dataset is the QCD background. Each dataset contains 14 rows and 100000 columns. 14
rows represent the 14 parameters we mentioned previously. They are ‘pt', 'eta', 'phi',
'mass', 'ee2', 'ee3', 'd2', 'angularity', 't1', 't2', 't3', 't21', 't32' , and 'KtDeltaR'. For each
parameter, we have recorded 100000 datapoints.
13
TRANSVERSE MOMENTUM(PT) (XY PLANE)(UNIT: GEV*C-1 OR
GEV)
The component of momentum in the transverse plane (in other words, perpendicular to
the beamline.) The parameter is always associated with the events that happened at the
vertex, which is a fairly clean sample.
Notice that the x-axis goes from 250 to 500. This agrees with the dataset we use where the
transverse momentum ranges from 250 GeV to 500 GeV. One peculiar thing to notice is that
there is a noticeable gap at the very end of the plot the Higgs datapoints peak around
485 GeV.
14
PSEUDORAPIDITY ETA (η)
Eta describes the angle of a particle relative to the beam axis(z-axis). θ and η are
convertible.
15
PHI (Φ)
16
INVARIANT MASS (M)(UNIT: GEV*C-2 OR GEV)
When a Higgs boson particle decays into other particles, its mass before the decay can be
calculated “from the energies and momenta of the decay products”(ATLAS). The mass is an
invariant quantity, which is the same for all observers in all reference frames.
We will plot the Higgs and QCD background mass and compare the two plots.
17
The difference between the Higgs and QCD data is fairly noticeable. Here, we are looking at
the “irregularities” of the plot, which would give us information about the signal.
18
EE2(E2) AND EE3(E3): ENERGY CORRELATION FUNCTIONS
19
More on the Energy Correlation Functions
Energy Correlations Functions can be used as an inquiry for the jet substructure, meaning
that we can use the functions to identify the N-prong substructure without a subject finding
procedure.
We can use ECFs to discriminate between QCD jets and Higgs Bosons.
A noticeable difference between the signal and background occurs when 2-point ECF is
greater than 0.11.
20
D2(D2) JET SHAPE FUNCTION, USED IN TAGGING
Note: The QCD background covers the Higgs data. The two datasets share similar
distributions.
21
ANGULARITY
Notes:
“Angularities are a family of observables that are sensitive to the degree of symmetry in the
energy flow inside a jet…..The measurement is aimed primarily at testing QCD, which
makes predictions for the shape of the angularity distribution in jets where the small-angle
approximation is valid…….Angularity is largely uncorrelated with all the other
observables.” (source: https://arxiv.org/pdf/1206.5369.pdf)
The Higgs and QCD data follow similar distributions.
22
N-SUBJETTINESS(𝜏N); JET SHAPE FUNCTION, USED IN TAGGING
23
More on Subjettiness.
The variable “describes to what degree the substructure of a given jet J is compatible with
being composed of N or fewer subjects.”
Tau 21, the ratio of the N-subjecttiness functions associated with the standard subject axes,
is used to generate the dimensionless variables that have shown to be particularly useful in
identifying two-body structures within jets. (ATLAS Collaboration)
24
KTDELTAR(KT ΔR) – KT CLUSTERING
25
Higgs Sensitivity
26
27
Conclusions
1. The most noticeable and practical result we can obtain from analyzing the datasets is
that the potential Higgs boson signal has an invariant mass of approximately 125 GeV.
2. It is observed that both the energy correlation function(ECF) and the N-subjettiness (𝜏)
can be used as a probe for the jet substructure. One parameter could be better than the
other to be used depending on the situation.
Energy Correlation functions: In the d2 parameter, the background and the signal share the
same distribution(Poisson-like). It seems like a good point where we apply a statistical test
to test the significance of a given candidate signal.
N-subjettiness: 𝜏21, the ratio, showed a noticeable separation between the background and
the Higgs signal.
3. It seems that it would be an effective way to partially separate the Higgs data from the
QCD background to increase the signal detection accuracy and reduce the background
noise by applying some constraints or thresholds depending on the parameters.
4. The ECF and the Subjettiness are the two most crucial parameters to discriminate
particles. Further detailed statistical tests and analysis should be done based on the two
parameters by truncating the datapoints or applying constraints.
28
Bibliography
“The Large Hadron Collider | CERN.”
Home.Cern
, 2019,
home.cern/science/accelerators/large-hadron-collider.
Sirunyan, A. M., et al. “Inclusive Search for a Highly Boosted Higgs Boson Decaying to a
Bottom Quark-Antiquark Pair.”
Physical Review Letters
, vol. 120, no. 7, 14 Feb. 2018,
arxiv.org/abs/1709.05543, 10.1103/physrevlett.120.071802. Accessed 6 Nov. 2019.
“Higgs Boson Observed Decaying to b Quarks – at Last! | ATLAS Experiment at CERN.”
ATLAS Experiment at CERN
, 9 July 2018, atlas.cern/updates/physics-briefing/higgs-
observed-decaying-b-quarks. Accessed 7 Nov. 2019.
“Mass / Invariant Mass | ATLAS Experiment at CERN.”
ATLAS Experiment at CERN
, 25 Jan.
2018, atlas.cern/glossary/mass-invariant-mass. Accessed 9 Nov. 2019.
Aad, G, et al. “Performance of Jet Substructure Techniques for Large-R Jets in Proton-Proton
Collisions at √s = 7 TeV Using the ATLAS Detector.”
Mit.Edu
, July 2013,
dspace.mit.edu/handle/1721.1/84505, 1029-8479. Accessed 10 Nov. 2019.
Aad, G., et al. “Identification of Boosted, Hadronically Decaying W Bosons and Comparisons
with ATLAS Data Taken at $$\sqrt{s} = 8$$ s = 8 TeV.”
The European Physical Journal C
,
vol. 76, no. 3, Mar. 2016, 10.1140/epjc/s10052-016-3978-z. Accessed 9 Nov. 2019.